Learning RIS-Assisted Hybrid NOMA Transmissions for Capacity Maximization in Wireless-Powered Networks

In this paper, we aim to enhance the throughput of a reconfigurable intelligent surface (RIS)-assisted and wireless-powered network by enabling user devices (UDs) to cooperatively adapt their channel access strategies. The base station (BS) simultaneously receives data and provides radio frequency (RF) energy to the UDs. Each UD equipped with a passive RIS employs non-orthogonal multiple access (NOMA) or backscattering for data transmissions. Moreover, the UD can assist other UDs' data transmissions through RIS's phase reconfiguration. We seek to maximize the overall throughput by optimizing UDs' channel access modes, time allocations, and the beamforming strategies of the RISs and the BS. Throughput maximization is challenging due to uncertain channel information, dynamic traffic demands, and limited energy supply. We propose a hierarchical learning approach to address access mode selection and transmission control in two steps. We first employ a multi-agent attention-enhanced deep reinforcement learning approach to update each UD's channel access strategy by interacting with the dynamic network. Then, we efficiently update the UDs' time allocation and the beamforming strategies of the RISs and the BS to further enhance throughput. Simulation results validate that our proposed scheme achieves improved throughput and learning efficiency, and exhibits superior adaptability against the varying traffic demands, compared to benchmarks.